Download PDFOpen PDF in browserFeature Engineering for development of a Machine Learning Model for Clash Resolution9 pages•Published: May 15, 2022AbstractTo automate clash resolution tasks, it is important to capture domain knowledge for the Machine Learning (ML). One way to add domain knowledge is by training data that divides tasks into input and output variables. The selection of input variables that are most relevant to a task is an important step towards automation. In this paper, the authors detail framework that uses literature review, industry interviews, and Modified Delphi to capture domain knowledge for clash resolution. The features identified through this paper can in future be processed through Feature Selection, that can provide empirical evidence of why the selected features or set of features are important to ML algorithm. Data collection processes discussed in this paper is not finalized and is discussed to help provide readers with framework of the proposed systematic method. Factors considered when resolving clashes were identified through literature review (22 factors) and industry interviews (16 factors). 14 factors identified from the interviews had a similar matching factor in the literature reviewed, the other 2 factors were not mentioned in any publications found during the initial literature review. After comparing results from literature review and interviews, 13 factors were considered critical for automating clash resolution.Keyphrases: clash coordination, clash resolution, dimensionality reduction, feature engineering, machine learning In: Tom Leathem, Wes Collins and Anthony Perrenoud (editors). ASC2022. 58th Annual Associated Schools of Construction International Conference, vol 3, pages 398-406.
|
